Update OCR to use tesseract 4.0.0

[Jason Morse]

Tesseract has been updated to 4.0.0 as of October 29, 2018.

https://github.com/tesseract-ocr/tesseract

The Syncfusion OCR library currently uses 3.02/5. I've found 4.0.0 to be much better for OCR of PDF documents. Is there a work item for upgrading OCR library to use newer tesseract library?

[Dilli Babu Nandha Gopal]

Hi Jason, 

 

We have tested the new Tesseract version and found that new Tesseract is performing slower than older version please find the details in the below table.  

Tesseract Version 4.0 OCR Process Time Taken Table Report (Syncfusion Tesseract .dll): 

  

Document	Size	Page count	TesseractVersion4.0 LSTM Engine   (OCR Time)	Tesseract Version 4.0  ( Tesseract Engine)  (OCR  Time )	Tesseract Version 3.05  (OCR Time)
Input.pdf	1.8 Mb	1	00.27.604 sec	00.13.496 sec	00.12.870 sec
Defect_143275.pdf	106 KB	1	00.19.612 sec	00.25.264 sec	00.18.931 sec
DefectID_WF11781.pdf	3.0 MB	8	00.58.832 sec	00.42.624 sec	00.44.559 sec
SpecialCharacters.pdf	15.7 KB	1	00.09.172 sec	00.07.523 sec	00.09.441 sec
DefectID_WF13618_1.pdf	3.6 MB	2	00.23.752 sec	00.37.851 sec	00.34.297 sec
WF25811.pdf	24.8 MB	15	04.59.782 sec	03.41.819 sec	03.50.944 sec
DefectID_WF32606.pdf	4.27 MB	12	01.16.325 sec	01.01.061 sec	1.04.171 sec
Defect_139301.pdf	5.4MB	32	05.14.944 sec	04.26.121 sec	04.18.701 sec

 

  

At present, we don't have any immediate plans provide support for this newer version. We have logged the feature request to this feature. We will let you know once this feature has been implemented. 

 

Regards, 

Dilli babu. 

 

 

[Jason Morse]

Thank you for the update. Although the current tesseract 4.0 performance generally is slower it is not my primary concern - recognition quality is. I am more than willing to consider taking a degradation in performance to achieve an much better improvement in recognition with LSTM engine.  

Document	Size	Page count	TesseractVersion4.0 LSTM Engine		Tesseract Version 4.0 ( Tesseract Engine)		Tesseract Version 3.05 (Baseline)
			(OCR Time sec)	Perf Improvement	(OCR Time sec)	Perf Improvement	(OCR Time sec)
Input.pdf	1.8 Mb	1	27.604	-114%	13.496	-5%	12.87
Defect_143275.pdf	106 KB	1	19.612	-4%	25.264	-33%	18.931
DefectID_WF11781.pdf	3.0 MB	8	58.832	-32%	42.624	4%	44.559
SpecialCharacters.pdf	15.7 KB	1	9.172	3%	7.523	20%	9.441
DefectID_WF13618_1.pdf	3.6 MB	2	23.752	31%	37.851	-10%	34.297
WF25811.pdf	24.8 MB	15	299.782	-30%	221.819	4%	230.944
DefectID_WF32606.pdf	4.27 MB	12	76.325	-19%	61.061	5%	64.171
Defect_139301.pdf	5.4MB	32	314.944	-22%	266.121	-3%	258.701

[Dilli Babu Nandha Gopal]

Hi Jason, 

 

Thank you for your update. 

  

We have considered your request and logged the feature request to this feature. We will implement this feature in any of our upcoming releases. The feature implementation would also greatly depend on the factors such as product design, code compatibility and complexity. We request you to visit our website periodically for feature related updates. 

 

Regards, 

Dilli babu. 

 

[Effy]

+1 on this request,

We've started using Tesseract as an external process instead of using SyncFusion due to the older version.

[Sowmiya Loganathan]

Hi Effy, 

 

At present the feature for “Update PDF OCR to Tesseract version 4.0” is not implemented. We will let you know once the feature is implemented. We request you to visit our website periodically for feature related updates.  

 

Regards, 

Sowmiya L 

[aJeff]

I also would very much like to have version 4.0 incorporated - the improved pre-processing and OCR output is well worth the slight performance hit

[Sowmiya Loganathan]

Hi aJeff, 

 

Thank you for the update. At present we do not have any immediate plans to “Update PDF OCR to Tesseract version 4.0”. Please visit our website periodically for feature related updates. 

 

Regards, 

Sowmiya L 

[Daniel Persidok]

Hello!

We're also interested in the 4.0.

The current version (3.05) is not even able to read simple numbers correct like the tax id in invoices.

BR

Daniel

[Surya Kumar]

Hi Daniel, 

We have logged the feature request for using Tesseract 4.0 in OCR and you can track the status of this feature from below link: 

https://www.syncfusion.com/feedback/5210/support-for-tesseract-4-0-in-syncfusion-ocr-processor 

We will let you know once this feature is implemented. 

Regards, 

Surya Kumar